EP3475879A1 - Domain separation neural networks - Google Patents

Domain separation neural networks

Info

Publication number
EP3475879A1
EP3475879A1 EP16757929.1A EP16757929A EP3475879A1 EP 3475879 A1 EP3475879 A1 EP 3475879A1 EP 16757929 A EP16757929 A EP 16757929A EP 3475879 A1 EP3475879 A1 EP 3475879A1
Authority
EP
European Patent Office
Prior art keywords
neural network
shared
images
input image
domain
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
EP16757929.1A
Other languages
German (de)
French (fr)
Other versions
EP3475879B1 (en
Inventor
Konstantinos BOUSMALIS
Nathan Silberman
Dilip Krishnan
George TRIGEORGIS
Dumitru Erhan
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Google LLC
Original Assignee
Google LLC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Google LLC filed Critical Google LLC
Publication of EP3475879A1 publication Critical patent/EP3475879A1/en
Application granted granted Critical
Publication of EP3475879B1 publication Critical patent/EP3475879B1/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/0895—Weakly supervised learning, e.g. semi-supervised or self-supervised learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/094—Adversarial learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/096—Transfer learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • G06T7/74—Determining position or orientation of objects or cameras using feature-based methods involving reference images or patches
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/72—Data preparation, e.g. statistical preprocessing of image or video features
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/7715—Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Definitions

  • This specification relates to processing image data through the layers of neural networks to generate outputs.
  • Neural networks are machine learning models that employ one or more layers of nonlinear units to predict an output for a received input.
  • Some neural networks include one or more hidden layers in addition to an output layer. The output of each hidden layer is used as input to the next layer in the network, i.e., the next hidden layer or the output layer.
  • Each layer of the network generates an output from a received input in accordance with current values of a respective set of parameters.
  • this specification describes image processing neural networks that are implemented by one or more computers and that generate network outputs for images from a target domain that characterize the images.
  • an image processing neural network system receives an input image from the target domain and processes the image to generate a shared feature representation of features of the input image that are shared between images from the target domain and images from a source domain different from the target domain. The image processing neural network system then generates a network output that characterizes the input image from the shared feature representation.
  • a system of one or more computers to be configured to perform particular operations or actions means that the system has installed on it software, firmware, hardware, or a combination of them that in operation cause the system to perform the operations or actions.
  • one or more computer programs to be configured to perform particular operations or actions means that the one or more programs include instructions that, when executed by data processing apparatus, cause the apparatus to perform the operations or actions.
  • the neural network system as described in this specification can outperform the state-of-the-art on a range of unsupervised domain adaptation scenarios and additionally produce visualizations of the private and shared representations that enable interpretation of the domain adaptation process during training.
  • the neural network system can learn representations that are domain invariant in scenarios where the data distributions during training and testing are different.
  • FIG. 1 shows an example image processing neural network system.
  • FIG. 2 shows an example neural network training system.
  • FIG. 3 is a flow diagram of an example process for generating a network output for an image from the target domain.
  • FIG. 4 is a flow diagram of an example process for performing an iteration of training a shared encoder neural network and a classifier neural network.
  • This specification describes an image processing neural network system
  • the image processing neural network system processes input images from a target domain to generate a respective network output for each input image that characterizes the network image.
  • the image processing neural network system can be configured to generate a network output that characterizes a received image from the target domain in any of a variety of ways.
  • the image processing neural network system can be configured to classify input images from the target domain as including images of objects from one or more object categories, i.e., to generate a classification output for an input image that includes a respective score for each of the one or more object categories that represents a likelihood that the input image includes an image of an object that belongs to the object category.
  • the image processing neural network system can be configured to estimate the pose of objects in input images.
  • the pose of an object is a combination of the position and orientation of the object in the input image.
  • the system can generate as the network output a pose vector that includes an estimated location in the image of each of a predetermined number of body joints of the object.
  • the image processing neural network system can be configured to divide input images into multiple regions, each of which belongs to a different one of a set of categories.
  • the system can generate, as the network output, data, e.g., a vector or a matrix, that identifies for each pixel of the input image the category to which the pixel belongs.
  • FIG. 1 shows an example image processing neural network system 100.
  • the image processing neural network system 100 is an example of a system implemented as computer programs on one or more computers in one or more locations, in which the systems, components, and techniques described below can be implemented.
  • the image processing neural network system 100 receives as input a target domain image 102 and processes the target domain image 102 to generate a network output 122. As described above, the image processing neural network system 100 can be configured to generate various kinds of network outputs for received input images.
  • the image processing neural network system 100 includes a shared encoder neural network 110 and a classifier neural network 120.
  • the shared encoder neural network 110 is a neural network, e.g., a convolutional neural network, that has been configured through training to receive the target domain image 102 and to process the target domain image 102 to generate a shared feature representation 112 for the target domain image 102.
  • the shared feature representation 112 is a vector of numeric values and is a representation of the features of the target domain image 102 that are shared between images from the target domain and images from a source domain.
  • the source domain is different from the target domain.
  • the distribution of pixel values in images from the source domain is different from the distribution of pixel values in images from the target domain.
  • the source and target domains differ primarily in terms of the distribution of low-level image statistics rather than high-level image statistics. Examples of low-level differences in the distributions include those arising due to noise, resolution, illumination and color. High-level differences, on the other hand, may include differences that relate to the number of classes, the types of objects, and geometric variations, such as 3D position.
  • the source domain may be images of a virtual environment that simulates a real- world environment and the target domain may be images of the real-world environment.
  • the source domain images may be images of a virtual environment that simulates a real-world environment that is to be interacted with by a robot or other mechanical agent or by an autonomous or semi-autonomous vehicle
  • the target domain images may be images of the real-world environment as captured by the mechanical agent or the vehicle.
  • the source domain images may be more expansive, in particular for edge cases that are occur rarely or are difficult or dangerous to generate in the real-world environment.
  • the source and target domains may both be real- world images, but with photometric differences, e.g., one domain may be indoor images and the other domain may be outdoor images of similar objects.
  • the source and target domains may be images selected from different collections of images of similar objects.
  • the source domain may be images from a curated image collection and the target domain may be images from a non- curated data set, e.g., Internet images.
  • the source domain may be a domain in which labeled images are easily available, while obtaining labeled image data for the target domain is computationally expensive or otherwise difficult or infeasible.
  • the classifier neural network 120 is a neural network, e.g., a fully-connected neural network, that has been configured through training to receive the shared feature representation 112 and to process the shared feature representation 112 to generate the network output 122 for the target domain image 102.
  • FIG. 2 shows an example neural network training system 200.
  • the neural network training system 200 is an example of a system implemented as computer programs on one or more computers in one or more locations, in which the systems, components, and techniques described below can be implemented.
  • the neural network training system 200 trains the shared encoder neural network 110 and the classifier neural network 120 to generate network outputs for received images from the target domain.
  • the neural network training system 200 trains the shared encoder neural network 110 on labeled images from the source domain and unlabeled images from the target domain and trains the classifier neural network 120 only on the labeled images from the source domain.
  • a labeled image is an image for which a known network output, i.e., the network output that should be generated by the classifier neural network 120 for the image, is available to the neural network training system 200 during training and an unlabeled image is an image for which no known network output is available to the neural network training system 200 during training.
  • the neural network training system 200 also includes a private target encoder neural network 210, a private source encoder neural network 220, and a shared decoder neural network 230.
  • the private target encoder neural network 210 is specific to the target domain and is configured to receive images from the target domain and to generate, for each received image, a private feature representation of the image.
  • the private source encoder neural network 220 is specific to the source domain and is configured to receive images from the source domain and to generate, for each received image, a private feature representation of the image.
  • the private target encoder neural network 210, the private source encoder neural network 220, and the shared encoder neural network 110 have the same neural network architecture, e.g., are all convolutional neural networks with the same network architecture.
  • the private target encoder neural network 210 and the private source encoder neural network 220 are convolutional neural networks with an architecture that differs from that of the shared encoder neural network 110.
  • the shared decoder neural network 230 is configured to receive a combined feature representation of an image and to process the combined feature representation to generate a reconstruction of the image.
  • the shared decoder neural network 230 is a deconvolutional neural network.
  • the combined feature representation of an image is a combination, e.g., a
  • the combined feature representation is a combination of the private feature representation generated for the image by the private source encoder neural network 220 and the shared feature representation generated for the image by the shared encoder neural network 110. If the image is from the target domain, the combined feature representation is a combination of the private feature representation generated for the image by the private target encoder neural network 210 and the shared feature representation generated for the image by the shared encoder neural network 1 10.
  • the neural network training system 200 adjusts the values of the parameters of the shared encoder neural network 1 10, the classifier neural network 120, the private target encoder neural network 210, the private source encoder neural network 220, and the shared decoder neural network 230 to determine trained values of the parameters of each of the networks.
  • the neural network training system 200 trains the shared encoder neural network 110 to (i) generate shared feature representations for input images from the target domain that are similar to shared feature representations for input images from the source domain while (ii) generating shared feature representations for input images from the target domain that are different from private feature representations for the same input images from the target domain generated by the private target encoder neural network 210 and (iii) generating shared feature representations for input images from the source domain that are different from private feature representations for the same input images from the source domain generated by the private source encoder neural network 220.
  • the neural network training system 200 trains the shared encoder neural network 110 to generate a shared feature representation for an input image from the target domain that, when combined with a private feature representation for the same input image generated by the private target encoder neural network 210, can be used to accurately reconstruct the input image by the shared decoder neural network 230 and to generate a shared feature representation for an input image from the source domain that, when combined with a private feature representation for the same input image generated by the private source encoder neural network 220, can be used to accurately reconstruct the input image by the shared decoder neural network 230.
  • the neural network training system 200 trains the classifier neural network 120 to generate accurate network outputs on the labeled images from the source domain.
  • the shared encoder neural network 110 and the shared decoder neural network 230 can be used to transform a target domain image into an image in the source domain.
  • the target domain image can be processed using the shared encoder neural network 110 to generate a shared representation.
  • the shared representation could then be combined with a placeholder or desired private feature representation generated for the image generated by the private source encoder neural network 220 to generate a combined representation.
  • the combined representation could then be processed by the shared decoder neural network 230 to generate a source domain image having the same features as the input target domain image.
  • FIG. 3 is a flow diagram of an example process 300 for generating a network output for an image from the target domain.
  • the process 300 will be described as being performed by a system of one or more computers located in one or more locations.
  • an image processing neural network system e.g., the image processing neural network system 100 of FIG.1, appropriately programmed, can perform the process 300.
  • the system receives an image from the target domain (step 302).
  • the system processes the target domain image using a shared encoder neural network (step 304).
  • the shared encoder neural network has been configured through training to receive the target domain image and to generate a shared feature representation of features of the target domain image that are shared between images from the target domain and images from the source domain.
  • the system processes the shared feature representation using a classifier neural network (step 306).
  • the classifier neural network has been configured through training to receive the shared feature representation and to process the shared feature representation to generate a network output for the target domain image.
  • FIG. 4 is a flow diagram of an example process 400 for performing an iteration of training a shared encoder neural network and a classifier neural network.
  • the process 400 will be described as being performed by a system of one or more computers located in one or more locations.
  • a neural network training system e.g., the neural network training system 200 of FIG.1, appropriately programmed, can perform the process 400.
  • the system receives labeled training source domain images and unlabeled training target domain images (step 402).
  • the system processes each training source domain image and each training target domain image using a shared encoder neural network in accordance with current values of the parameters of the shared encoder neural network to generate a respective shared feature representation for each of the images (step 404).
  • the system processes each training target domain image using a private target encoder neural network in accordance with current values of the parameters of the private target encoder neural network to generate a respective private feature representation for each training target domain image (step 406).
  • the system processes each training source domain image using a private source encoder neural network in accordance with current values of the parameters of the private source encoder neural network to generate a respective private feature representation for each training source domain image (step 408).
  • the system For each training source domain image, the system processes the shared feature representation for the training source domain image using a classifier neural network in accordance with current values of the parameters of the classifier neural network to generate a network output for the training source domain image (step 410).
  • the system generates a respective combined representation for each training source domain image and each training target domain image, e.g., as described above with reference to FIG. 2 (step 412).
  • the system processes the combined representation for each of the training images using a shared decoder neural network in accordance with current values of the parameters of the shared decoder neural network to generate a reconstruction of the corresponding training image (step 414).
  • the system adjusts the current values of the parameters of the private target encoder neural network, the shared encoder neural network, the private source encoder neural network, the shared decoder neural network, and the classifier neural network to reduce a loss resulting from the outputs generated by the neural networks (step 416).
  • the system performs an iteration of a machine learning training technique, e.g., stochastic gradient with momentum, to minimize a loss that depends on a classification loss, a reconstruction loss, a difference loss, and a similarity loss.
  • a machine learning training technique e.g., stochastic gradient with momentum
  • the loss ma satisfy: where is the classification loss, is the reconstruction loss, is the difference loss, is the simirality loss and ⁇ , ⁇ , and ⁇ are weights that control the interaction of the loss terms.
  • the classification loss trains the classifier neural network and, by virtue of backpropogation, the shared encoder neural network to generate accurate network outputs for source domain images, i.e., to generate network outputs that match the known network otuputs for the training source domain images.
  • the classification loss may satisfy: where N s is the total number of labeled training source domain images, is the network output generated by the classifier neural network for the z-th training source domain image, and is the known network output for the z ' -th training source domain image.
  • the reconstruction loss trains the shared decoder neural network and, by virtue of backpropogation, each of the encoder neural networks to generate accurate reconstructions of input images, i.e., to generate reconstructions of the training images that match the training images.
  • the reconstruction loss may be a scale-invariant mean squared error loss that satisfies:
  • N is the total number of training target domain images
  • the scale-invariant mean squared error penalizes differences between pairs of pixels, i.e., a pixel in a given training image and the corresponding pixel in the reconstruction of the training image.
  • the system trains the shared decoder neural network to reproduce the overall shape of the objects being modeled without expending modeling power on the absolute color or intensity of the inputs.
  • the difference loss trains the shared encoder neural network to (i) generate shared feature representations for input images from the target domain that are different from private feature representations for the same input images from the target domain generated by the private target encoder neural network and (ii) generate shared feature representations for input images from the source domain that are different from private feature representations for the same input images from the source domain generated by the private source encoder neural network.
  • the difference loss may satisfy:
  • the similarity loss trains the shared encoder neural network to generate shared feature representations for input images from the target domain that are similar to shared feature representations for input images from the source domain.
  • the similarity loss may be a domain adversarial similarity loss that trains the shared encoder neural network to generate the shared representations such that a domain classifier neural network cannot reliably predict the domain of the encoded representation.
  • the domain adversarial similarity loss may satisfy:
  • di is the value of a variable that identifies whether the 2-th training image is from the target or source domain, and is the predicted value of the variable as predicted by the domain classifier neural network by processing the shared feature representation for the i-t training image.
  • the system can maximize this loss with respect to the parameters of the domain classifier neural network and minimize the loss with the respect to the parameters of the shared encoder neural network, i.e., by including a gradient reversal layer as the input layer of the domain classifier neural network and minimizing the overall loss.
  • the similarity loss may be a Maximum Mean Discrepancy loss that is a kernel-based distance function between pairs of samples.
  • PSD kernel function is the shared feature representation of a training source domain image, and is the shared feature representation of a training target domain image.
  • the PSD kernel function may be a linear combination of multiple different RBF kernels.
  • the system can repeatedly perform the process 400 on different batches of labeled training source domain images and unlabeled training target domain images to train the neural networks by iteratively adjusting the values of the parameters of the neural networks to determine trained values of the parameters of the neural networks.
  • Embodiments of the subject matter and the functional operations described in this specification can be implemented in digital electronic circuitry, in tangibly- embodied computer software or firmware, in computer hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them.
  • Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non transitory program carrier for execution by, or to control the operation of, data processing apparatus.
  • the program i.e., one or more modules of computer program instructions encoded on a tangible non transitory program carrier for execution by, or to control the operation of, data processing apparatus.
  • the program i.e., one or more modules of computer program instructions encoded on a tangible non transitory program carrier for execution by, or to control the operation of, data processing apparatus.
  • the program i.e., one or more modules of computer program instructions encoded on a tangible non transitory program carrier for execution by,
  • instructions can be encoded on an artificially generated propagated signal, e.g., a machine- generated electrical, optical, or electromagnetic signal, that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus.
  • the computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.
  • data processing apparatus encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable processor, a computer, or multiple processors or computers.
  • the apparatus can include special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit).
  • the apparatus can also include, in addition to hardware, code that creates an execution environment for the computer program in question, e.g., code that constitutes processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them.
  • a computer program (which may also be referred to or described as a program, software, a software application, a module, a software module, a script, or code) can be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.
  • a computer program may, but need not, correspond to a file in a file system.
  • a program can be stored in a portion of a file that holds other programs or data, e.g., one or more scripts stored in a markup language document, in a single file dedicated to the program in question, or in multiple coordinated files, e.g., files that store one or more modules, sub programs, or portions of code.
  • a computer program can be deployed to be executed on one computer or on multiple computers that are located at one site or distributed across multiple sites and interconnected by a communication network.
  • the processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data and generating output.
  • the processes and logic flows can also be performed by, and apparatus can also be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit).
  • special purpose logic circuitry e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit).
  • Computers suitable for the execution of a computer program include, by way of example, can be based on general or special purpose microprocessors or both, or any other kind of central processing unit.
  • a central processing unit will receive instructions and data from a read only memory or a random access memory or both.
  • the essential elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices for storing instructions and data.
  • a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto optical disks, or optical disks.
  • mass storage devices for storing data, e.g., magnetic, magneto optical disks, or optical disks.
  • a computer need not have such devices.
  • a computer can be embedded in another device, e.g., a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device, e.g., a universal serial bus (USB) flash drive, to name just a few.
  • Computer readable media suitable for storing computer program instructions and data include all forms of non- volatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto optical disks; and CD ROM and DVD-ROM disks.
  • the processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.
  • a computer having a display device, e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user and a keyboard and a pointing device, e.g., a mouse or a trackball, by which the user can provide input to the computer.
  • a display device e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor
  • keyboard and a pointing device e.g., a mouse or a trackball
  • Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input.
  • a computer can interact with a user by sending documents to and receiving documents from a device that is used by the user; for example, by sending web pages to a
  • Embodiments of the subject matter described in this specification can be implemented in a computing system that includes a back end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front end component, e.g., a client computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the subject matter described in this specification, or any combination of one or more such back end, middleware, or front end components.
  • the components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network (“LAN”) and a wide area network (“WAN”), e.g., the Internet.
  • LAN local area network
  • WAN wide area network
  • the computing system can include clients and servers.
  • a client and server are generally remote from each other and typically interact through a communication network.
  • the relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Multimedia (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images using an image processing neural network system. One of the system includes a shared encoder neural network implemented by one or more computers, wherein the shared encoder neural network is configured to: receive an input image from a target domain; and process the input image to generate a shared feature representation of features of the input image that are shared between images from the target domain and images from a source domain different from the target domain; and a classifier neural network implemented by the one or more computers, wherein the classifier neural network is configured to: receive the shared feature representation; and process the shared feature representation to generate a network output for the input image that characterizes the input image.

Description

DOMAIN SEPARATION NEURAL NETWORKS
BACKGROUND
This specification relates to processing image data through the layers of neural networks to generate outputs.
Neural networks are machine learning models that employ one or more layers of nonlinear units to predict an output for a received input. Some neural networks include one or more hidden layers in addition to an output layer. The output of each hidden layer is used as input to the next layer in the network, i.e., the next hidden layer or the output layer. Each layer of the network generates an output from a received input in accordance with current values of a respective set of parameters.
SUMMARY
In general, this specification describes image processing neural networks that are implemented by one or more computers and that generate network outputs for images from a target domain that characterize the images. In particular, an image processing neural network system receives an input image from the target domain and processes the image to generate a shared feature representation of features of the input image that are shared between images from the target domain and images from a source domain different from the target domain. The image processing neural network system then generates a network output that characterizes the input image from the shared feature representation.
For a system of one or more computers to be configured to perform particular operations or actions means that the system has installed on it software, firmware, hardware, or a combination of them that in operation cause the system to perform the operations or actions. For one or more computer programs to be configured to perform particular operations or actions means that the one or more programs include instructions that, when executed by data processing apparatus, cause the apparatus to perform the operations or actions.
Particular embodiments of the subject matter described in this specification can be implemented so as to realize one or more of the following advantages. The neural network system as described in this specification can outperform the state-of-the-art on a range of unsupervised domain adaptation scenarios and additionally produce visualizations of the private and shared representations that enable interpretation of the domain adaptation process during training. The neural network system can learn representations that are domain invariant in scenarios where the data distributions during training and testing are different.
The details of one or more embodiments of the subject matter of this specification are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages of the subject matter will become apparent from the description, the drawings, and the claims.
BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 shows an example image processing neural network system.
FIG. 2 shows an example neural network training system.
FIG. 3 is a flow diagram of an example process for generating a network output for an image from the target domain.
FIG. 4 is a flow diagram of an example process for performing an iteration of training a shared encoder neural network and a classifier neural network.
Like reference numbers and designations in the various drawings indicate like elements.
DETAILED DESCRIPTION
This specification describes an image processing neural network system
implemented as computer programs on one or more computers in one or more locations.
The image processing neural network system processes input images from a target domain to generate a respective network output for each input image that characterizes the network image.
The image processing neural network system can be configured to generate a network output that characterizes a received image from the target domain in any of a variety of ways.
For example, the image processing neural network system can be configured to classify input images from the target domain as including images of objects from one or more object categories, i.e., to generate a classification output for an input image that includes a respective score for each of the one or more object categories that represents a likelihood that the input image includes an image of an object that belongs to the object category.
As another example, the image processing neural network system can be configured to estimate the pose of objects in input images. Generally, the pose of an object is a combination of the position and orientation of the object in the input image. For example, the system can generate as the network output a pose vector that includes an estimated location in the image of each of a predetermined number of body joints of the object.
As another example, the image processing neural network system can be configured to divide input images into multiple regions, each of which belongs to a different one of a set of categories. For example, the system can generate, as the network output, data, e.g., a vector or a matrix, that identifies for each pixel of the input image the category to which the pixel belongs.
FIG. 1 shows an example image processing neural network system 100. The image processing neural network system 100 is an example of a system implemented as computer programs on one or more computers in one or more locations, in which the systems, components, and techniques described below can be implemented.
The image processing neural network system 100 receives as input a target domain image 102 and processes the target domain image 102 to generate a network output 122. As described above, the image processing neural network system 100 can be configured to generate various kinds of network outputs for received input images.
The image processing neural network system 100 includes a shared encoder neural network 110 and a classifier neural network 120.
The shared encoder neural network 110 is a neural network, e.g., a convolutional neural network, that has been configured through training to receive the target domain image 102 and to process the target domain image 102 to generate a shared feature representation 112 for the target domain image 102. The shared feature representation 112 is a vector of numeric values and is a representation of the features of the target domain image 102 that are shared between images from the target domain and images from a source domain.
Generally, the source domain is different from the target domain. In particular, the distribution of pixel values in images from the source domain is different from the distribution of pixel values in images from the target domain. In some cases, the source and target domains differ primarily in terms of the distribution of low-level image statistics rather than high-level image statistics. Examples of low-level differences in the distributions include those arising due to noise, resolution, illumination and color. High-level differences, on the other hand, may include differences that relate to the number of classes, the types of objects, and geometric variations, such as 3D position.
In one example, the source domain may be images of a virtual environment that simulates a real- world environment and the target domain may be images of the real-world environment. For example, the source domain images may be images of a virtual environment that simulates a real-world environment that is to be interacted with by a robot or other mechanical agent or by an autonomous or semi-autonomous vehicle, while the target domain images may be images of the real-world environment as captured by the mechanical agent or the vehicle. Thus, the source domain images may be more expansive, in particular for edge cases that are occur rarely or are difficult or dangerous to generate in the real-world environment. By training the system to generate accurate network outputs for target domain images using only labeled source domain images as described in this specification, the performance of the agent or vehicle in the real- world environment may be improved, even if certain situations are only encountered in the simulated version.
In another example, the source and target domains may both be real- world images, but with photometric differences, e.g., one domain may be indoor images and the other domain may be outdoor images of similar objects.
In another example, the source and target domains may be images selected from different collections of images of similar objects. For example, the source domain may be images from a curated image collection and the target domain may be images from a non- curated data set, e.g., Internet images.
In some cases, the source domain may be a domain in which labeled images are easily available, while obtaining labeled image data for the target domain is computationally expensive or otherwise difficult or infeasible.
The classifier neural network 120 is a neural network, e.g., a fully-connected neural network, that has been configured through training to receive the shared feature representation 112 and to process the shared feature representation 112 to generate the network output 122 for the target domain image 102.
Training the shared encoder neural network 110 and the classifier neural network 120 to generate the shared feature representation 112 and the network output 122 is described in more detail below with reference to FIGS. 2 and 4.
FIG. 2 shows an example neural network training system 200. The neural network training system 200 is an example of a system implemented as computer programs on one or more computers in one or more locations, in which the systems, components, and techniques described below can be implemented.
The neural network training system 200 trains the shared encoder neural network 110 and the classifier neural network 120 to generate network outputs for received images from the target domain.
In particular, the neural network training system 200 trains the shared encoder neural network 110 on labeled images from the source domain and unlabeled images from the target domain and trains the classifier neural network 120 only on the labeled images from the source domain. A labeled image is an image for which a known network output, i.e., the network output that should be generated by the classifier neural network 120 for the image, is available to the neural network training system 200 during training and an unlabeled image is an image for which no known network output is available to the neural network training system 200 during training.
To train the shared encoder neural network 1 10 and the classifier neural network 120, the neural network training system 200 also includes a private target encoder neural network 210, a private source encoder neural network 220, and a shared decoder neural network 230.
The private target encoder neural network 210 is specific to the target domain and is configured to receive images from the target domain and to generate, for each received image, a private feature representation of the image.
The private source encoder neural network 220 is specific to the source domain and is configured to receive images from the source domain and to generate, for each received image, a private feature representation of the image.
In some implementations, the private target encoder neural network 210, the private source encoder neural network 220, and the shared encoder neural network 110 have the same neural network architecture, e.g., are all convolutional neural networks with the same network architecture. In some other implementations, the private target encoder neural network 210 and the private source encoder neural network 220 are convolutional neural networks with an architecture that differs from that of the shared encoder neural network 110.
The shared decoder neural network 230 is configured to receive a combined feature representation of an image and to process the combined feature representation to generate a reconstruction of the image. In some implementations, the shared decoder neural network 230 is a deconvolutional neural network.
The combined feature representation of an image is a combination, e.g., a
concatenation, a sum, or an average, of the shared feature representation of the image and the private feature representation of the image. That is, if the image is from the source domain, the combined feature representation is a combination of the private feature representation generated for the image by the private source encoder neural network 220 and the shared feature representation generated for the image by the shared encoder neural network 110. If the image is from the target domain, the combined feature representation is a combination of the private feature representation generated for the image by the private target encoder neural network 210 and the shared feature representation generated for the image by the shared encoder neural network 1 10.
During the training, the neural network training system 200 adjusts the values of the parameters of the shared encoder neural network 1 10, the classifier neural network 120, the private target encoder neural network 210, the private source encoder neural network 220, and the shared decoder neural network 230 to determine trained values of the parameters of each of the networks.
In particular, the neural network training system 200 trains the shared encoder neural network 110 to (i) generate shared feature representations for input images from the target domain that are similar to shared feature representations for input images from the source domain while (ii) generating shared feature representations for input images from the target domain that are different from private feature representations for the same input images from the target domain generated by the private target encoder neural network 210 and (iii) generating shared feature representations for input images from the source domain that are different from private feature representations for the same input images from the source domain generated by the private source encoder neural network 220.
Additionally, the neural network training system 200 trains the shared encoder neural network 110 to generate a shared feature representation for an input image from the target domain that, when combined with a private feature representation for the same input image generated by the private target encoder neural network 210, can be used to accurately reconstruct the input image by the shared decoder neural network 230 and to generate a shared feature representation for an input image from the source domain that, when combined with a private feature representation for the same input image generated by the private source encoder neural network 220, can be used to accurately reconstruct the input image by the shared decoder neural network 230.
Further, as part of the training, the neural network training system 200 trains the classifier neural network 120 to generate accurate network outputs on the labeled images from the source domain.
Training the shared encoder neural network 110, the classifier neural network 120, the private target encoder neural network 210, the private source encoder neural network 220, and the shared decoder neural network 230 to determine trained values of the parameters of each of the networks is described in more detail below with reference to FIG. 4.
In some implementations, once trained, the shared encoder neural network 110 and the shared decoder neural network 230 can be used to transform a target domain image into an image in the source domain. For example, the target domain image can be processed using the shared encoder neural network 110 to generate a shared representation. The shared representation could then be combined with a placeholder or desired private feature representation generated for the image generated by the private source encoder neural network 220 to generate a combined representation. The combined representation could then be processed by the shared decoder neural network 230 to generate a source domain image having the same features as the input target domain image.
FIG. 3 is a flow diagram of an example process 300 for generating a network output for an image from the target domain. For convenience, the process 300 will be described as being performed by a system of one or more computers located in one or more locations. For example, an image processing neural network system, e.g., the image processing neural network system 100 of FIG.1, appropriately programmed, can perform the process 300.
The system receives an image from the target domain (step 302).
The system processes the target domain image using a shared encoder neural network (step 304). As described above, the shared encoder neural network has been configured through training to receive the target domain image and to generate a shared feature representation of features of the target domain image that are shared between images from the target domain and images from the source domain.
The system processes the shared feature representation using a classifier neural network (step 306). As described above, the classifier neural network has been configured through training to receive the shared feature representation and to process the shared feature representation to generate a network output for the target domain image.
FIG. 4 is a flow diagram of an example process 400 for performing an iteration of training a shared encoder neural network and a classifier neural network. For convenience, the process 400 will be described as being performed by a system of one or more computers located in one or more locations. For example, a neural network training system, e.g., the neural network training system 200 of FIG.1, appropriately programmed, can perform the process 400.
The system receives labeled training source domain images and unlabeled training target domain images (step 402).
The system processes each training source domain image and each training target domain image using a shared encoder neural network in accordance with current values of the parameters of the shared encoder neural network to generate a respective shared feature representation for each of the images (step 404).
The system processes each training target domain image using a private target encoder neural network in accordance with current values of the parameters of the private target encoder neural network to generate a respective private feature representation for each training target domain image (step 406).
The system processes each training source domain image using a private source encoder neural network in accordance with current values of the parameters of the private source encoder neural network to generate a respective private feature representation for each training source domain image (step 408).
For each training source domain image, the system processes the shared feature representation for the training source domain image using a classifier neural network in accordance with current values of the parameters of the classifier neural network to generate a network output for the training source domain image (step 410).
The system generates a respective combined representation for each training source domain image and each training target domain image, e.g., as described above with reference to FIG. 2 (step 412).
The system processes the combined representation for each of the training images using a shared decoder neural network in accordance with current values of the parameters of the shared decoder neural network to generate a reconstruction of the corresponding training image (step 414).
The system adjusts the current values of the parameters of the private target encoder neural network, the shared encoder neural network, the private source encoder neural network, the shared decoder neural network, and the classifier neural network to reduce a loss resulting from the outputs generated by the neural networks (step 416).
In particular, the system performs an iteration of a machine learning training technique, e.g., stochastic gradient with momentum, to minimize a loss that depends on a classification loss, a reconstruction loss, a difference loss, and a similarity loss. For example, the loss ma satisfy: where is the classification loss, is the reconstruction loss, is the difference loss, is the simirality loss and α, β, and γ are weights that control the interaction of the loss terms.
The classification loss trains the classifier neural network and, by virtue of backpropogation, the shared encoder neural network to generate accurate network outputs for source domain images, i.e., to generate network outputs that match the known network otuputs for the training source domain images. For example, the classification loss may satisfy: where Ns is the total number of labeled training source domain images, is the network output generated by the classifier neural network for the z-th training source domain image, and is the known network output for the z'-th training source domain image.
The reconstruction loss trains the shared decoder neural network and, by virtue of backpropogation, each of the encoder neural networks to generate accurate reconstructions of input images, i.e., to generate reconstructions of the training images that match the training images. For example, the reconstruction loss may be a scale-invariant mean squared error loss that satisfies:
where N, is the total number of training target domain images, is the z'-th training source domain image, is the reconstruction of the z'-th training source domain image generated by the shared decoder neural network, is the z'-th training target domain image, is the reconstruction of the z'-th training target domain image generated by the shared decoder neural network, and is the scale-invariant mean squared error between an image and the reconstruction of the image that satisfies where is the squared Z^-norm, k is the number of pixels in input x, and is a vector of ones of length k. Thus, the scale-invariant mean squared error penalizes differences between pairs of pixels, i.e., a pixel in a given training image and the corresponding pixel in the reconstruction of the training image. By minimizing this scale- invariant mean squared error loss, the system trains the shared decoder neural network to reproduce the overall shape of the objects being modeled without expending modeling power on the absolute color or intensity of the inputs.
The difference loss trains the shared encoder neural network to (i) generate shared feature representations for input images from the target domain that are different from private feature representations for the same input images from the target domain generated by the private target encoder neural network and (ii) generate shared feature representations for input images from the source domain that are different from private feature representations for the same input images from the source domain generated by the private source encoder neural network. For example, the difference loss may satisfy:
where is a matrix having rows that are the shared feature representations of the training source domain images, is a matrix having rows that are the shared feature representations of the training target domain images, is a matrix having rows that are the private feature representations of the training source domain images, is a matrix having rows that are the private feature representations of the training target domain images, and is the squared Frobenius norm.
The similarity loss trains the shared encoder neural network to generate shared feature representations for input images from the target domain that are similar to shared feature representations for input images from the source domain.
For example, in some cases, the similarity loss may be a domain adversarial similarity loss that trains the shared encoder neural network to generate the shared representations such that a domain classifier neural network cannot reliably predict the domain of the encoded representation. For example, the domain adversarial similarity loss may satisfy:
where di is the value of a variable that identifies whether the 2-th training image is from the target or source domain, and is the predicted value of the variable as predicted by the domain classifier neural network by processing the shared feature representation for the i-t training image. The system can maximize this loss with respect to the parameters of the domain classifier neural network and minimize the loss with the respect to the parameters of the shared encoder neural network, i.e., by including a gradient reversal layer as the input layer of the domain classifier neural network and minimizing the overall loss.
As another example, in some other cases, the similarity loss may be a Maximum Mean Discrepancy loss that is a kernel-based distance function between pairs of samples.
For example, the Maximum Mean Discrepancy loss may satisfy:
where is a PSD kernel function, is the shared feature representation of a training source domain image, and is the shared feature representation of a training target domain image. In some cases, the PSD kernel function may be a linear combination of multiple different RBF kernels.
The system can repeatedly perform the process 400 on different batches of labeled training source domain images and unlabeled training target domain images to train the neural networks by iteratively adjusting the values of the parameters of the neural networks to determine trained values of the parameters of the neural networks.
Embodiments of the subject matter and the functional operations described in this specification can be implemented in digital electronic circuitry, in tangibly- embodied computer software or firmware, in computer hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible non transitory program carrier for execution by, or to control the operation of, data processing apparatus. Alternatively or in addition, the program
instructions can be encoded on an artificially generated propagated signal, e.g., a machine- generated electrical, optical, or electromagnetic signal, that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus. The computer storage medium can be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.
The term "data processing apparatus" encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable processor, a computer, or multiple processors or computers. The apparatus can include special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit). The apparatus can also include, in addition to hardware, code that creates an execution environment for the computer program in question, e.g., code that constitutes processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them.
A computer program (which may also be referred to or described as a program, software, a software application, a module, a software module, a script, or code) can be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program may, but need not, correspond to a file in a file system. A program can be stored in a portion of a file that holds other programs or data, e.g., one or more scripts stored in a markup language document, in a single file dedicated to the program in question, or in multiple coordinated files, e.g., files that store one or more modules, sub programs, or portions of code. A computer program can be deployed to be executed on one computer or on multiple computers that are located at one site or distributed across multiple sites and interconnected by a communication network.
The processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform functions by operating on input data and generating output. The processes and logic flows can also be performed by, and apparatus can also be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit).
Computers suitable for the execution of a computer program include, by way of example, can be based on general or special purpose microprocessors or both, or any other kind of central processing unit. Generally, a central processing unit will receive instructions and data from a read only memory or a random access memory or both. The essential elements of a computer are a central processing unit for performing or executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic, magneto optical disks, or optical disks. However, a computer need not have such devices. Moreover, a computer can be embedded in another device, e.g., a mobile telephone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device, e.g., a universal serial bus (USB) flash drive, to name just a few. Computer readable media suitable for storing computer program instructions and data include all forms of non- volatile memory, media and memory devices, including by way of example semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto optical disks; and CD ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.
To provide for interaction with a user, embodiments of the subject matter described in this specification can be implemented on a computer having a display device, e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user and a keyboard and a pointing device, e.g., a mouse or a trackball, by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input. In addition, a computer can interact with a user by sending documents to and receiving documents from a device that is used by the user; for example, by sending web pages to a web browser on a user's client device in response to requests received from the web browser.
Embodiments of the subject matter described in this specification can be implemented in a computing system that includes a back end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front end component, e.g., a client computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the subject matter described in this specification, or any combination of one or more such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network ("LAN") and a wide area network ("WAN"), e.g., the Internet.
The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.
While this specification contains many specific implementation details, these should not be construed as limitations on the scope of any invention or of what may be claimed, but rather as descriptions of features that may be specific to particular embodiments of particular inventions. Certain features that are described in this specification in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination.
Moreover, although features may be described above as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.
Similarly, while operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing may be advantageous. Moreover, the separation of various system modules and components in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.
Particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. For example, the actions recited in the claims can be performed in a different order and still achieve desirable results; As one example, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In certain implementations, multitasking and parallel processing may be advantageous.
What is claimed is:

Claims

1. A system comprising:
a shared encoder neural network implemented by one or more computers, wherein the shared encoder neural network is configured to:
receive an input image from a target domain; and
process the input image to generate a shared feature representation of features of the input image that are shared between images from the target domain and images from a source domain different from the target domain; and
a classifier neural network implemented by the one or more computers, wherein the classifier neural network is configured to:
receive the shared feature representation; and
process the shared feature representation to generate a network output for the input image that characterizes the input image.
2. The system of claim 1, wherein images from the target domain have different low level image statistics than images from images from the source domain.
3. The system of any one of claims 1 or 2, wherein the network output is an object classification output.
4. The system of any one of claims 1 or 2, wherein the network output is a pose estimation output.
5. The system of any one of claims 1-4, wherein classifier neural network has been trained on labeled images from the source domain.
6. The system of any one of claims 1-5, wherein the shared encoder neural network has been trained to generate shared feature.representations for input images from the target domain that are similar to shared feature representations for input images from the source domain.
7. The system of claim 6, wherein the shared encoder neural network has been trained to generate shared feature representations for input images from the target domain that are different from private feature representations for the same input images from the target domain generated by a private target encoder neural network that is specific to the target domain.
8. The system of claim 7, wherein the shared encoder neural network has been trained to generate shared feature representations for input images from the source domain that are different from private feature representations for the same input images from the source domain generated by a private source encoder neural network that is specific to the source domain.
9. The system of claim 8, wherein the shared encoder neural network has been trained to generate a shared feature representation for an input image from the target domain that, when combined with a private feature representation for the same input image generated by the private target encoder neural network, can be used to accurately reconstruct the input image by a shared decoder neural network.
10. The system of claim 9, wherein the shared encoder neural network has been trained to generate a shared feature representation for an input image from the source domain that, when combined with a private feature representation for the same input image generated by the private source encoder neural network, can be used to accurately reconstruct the input image by the shared decoder neural network.
11. The system of claim 10, wherein the shared encoder neural network, the private source encoder neural network, and the private target encoder neural network are
convolutional neural networks with a same architecture but different parameter values.
12. The system of any one of claims 1-11, wherein the classifier neural network is a fully- connected neural network.
13. A computer storage medium encoded with instructions that, when executed by one or more computers, cause the one or more computers to implemented the system of any one of claims 1-12.
14. A method performed by one or more computers, the method comprising:
receiving an input image from a target domain;
processing the input image from the target domain using a shared encoder neural network, wherein the shared encoder neural network is configured to:
receive the input image from the target domain; and
process the input image to generate a shared feature representation of features of the input image that are shared between images from the target domain and images from a source domain different from the target domain; and
processing the input image from the target domain using a classifier neural network, wherein the classifier neural network is configured to:
receive the shared feature representation; and
process the shared feature representation to generate a network output for the input image that characterizes the input image.
15. The method of claim 14, wherein images from the target domain have different low level image statistics than images from images from the source domain.
16. The method of any one of claims 14 or 15, wherein the network output is an object classification output.
17. The method of any one of claims 14 or 15, wherein the network output is a pose estimation output.
18. The method of any one of claims 14-17, wherein classifier neural network has been trained on labeled images from the source domain.
19. The method of any one of claims 14-18, wherein the shared encoder neural network has been trained to generate shared feature representations for input images from the target domain that are similar to shared feature representations for input images from the source domain.
20. The method of claim 19, wherein the shared encoder neural network has been trained to generate shared feature representations for input images from the target domain that are different from private feature representations for the same input images from the target domain generated by a private target encoder neural network that is specific to the target domain.
21. The method of claim 20, wherein the shared encoder neural network has been trained to generate shared feature representations for input images from the source domain that are different from private feature representations for the same input images from the source domain generated by a private source encoder neural network that is specific to the source domain.
22. The method of claim 21 , wherein the shared encoder neural network has been trained to generate a shared feature representation for an input image from the target domain that, when combined with a private feature representation for the same input image generated by the private target encoder neural network, can be used to accurately reconstruct the input image by a shared decoder neural network.
23. The method of claim 22, wherein the shared encoder neural network has been trained to generate a shared feature representation for an input image from the source domain that, when combined with a private feature representation for the same input image generated by the private source encoder neural network, can be used to accurately reconstruct the input image by the shared decoder neural network.
24. The method of claim 23, wherein the shared encoder neural network, the private source encoder neural network, and the private target encoder neural network are convolutional neural networks with a same architecture but different parameter values.
25. The method of any one of claims 14-24, wherein the classifier neural network is a fully-connected neural network.
EP16757929.1A 2016-07-28 2016-07-28 Domain separation neural networks Active EP3475879B1 (en)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/GR2016/000035 WO2018020277A1 (en) 2016-07-28 2016-07-28 Domain separation neural networks

Publications (2)

Publication Number Publication Date
EP3475879A1 true EP3475879A1 (en) 2019-05-01
EP3475879B1 EP3475879B1 (en) 2025-03-19

Family

ID=56842965

Family Applications (1)

Application Number Title Priority Date Filing Date
EP16757929.1A Active EP3475879B1 (en) 2016-07-28 2016-07-28 Domain separation neural networks

Country Status (6)

Country Link
US (2) US10970589B2 (en)
EP (1) EP3475879B1 (en)
JP (1) JP6771645B2 (en)
KR (1) KR102318772B1 (en)
CN (1) CN109643383B (en)
WO (1) WO2018020277A1 (en)

Families Citing this family (56)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11177022B2 (en) 2016-10-17 2021-11-16 International Business Machines Corporation Workflow for automatic measurement of doppler pipeline
US20180107791A1 (en) * 2016-10-17 2018-04-19 International Business Machines Corporation Cohort detection from multimodal data and machine learning
US10529088B2 (en) 2016-12-02 2020-01-07 Gabriel Fine Automatically determining orientation and position of medically invasive devices via image processing
US10832166B2 (en) * 2016-12-20 2020-11-10 Conduent Business Services, Llc Method and system for text classification based on learning of transferable feature representations from a source domain
US11200459B1 (en) * 2017-10-10 2021-12-14 Snap Inc. Adversarial network for transfer learning
JP7013489B2 (en) * 2018-01-26 2022-02-15 株式会社ソニー・インタラクティブエンタテインメント Learning device, live-action image classification device generation system, live-action image classification device generation device, learning method and program
CN108460415B (en) * 2018-02-28 2021-06-15 国信优易数据股份有限公司 A method of language recognition
JP7063052B2 (en) 2018-03-28 2022-05-09 富士通株式会社 Goodness-of-fit calculation program, goodness-of-fit calculation method, goodness-of-fit calculation device, identification program, identification method and identification device
JP6893194B2 (en) * 2018-05-28 2021-06-23 日本電信電話株式会社 Model learning device, model learning method, and program
US10915792B2 (en) * 2018-09-06 2021-02-09 Nec Corporation Domain adaptation for instance detection and segmentation
US11640519B2 (en) * 2018-10-31 2023-05-02 Sony Interactive Entertainment Inc. Systems and methods for domain adaptation in neural networks using cross-domain batch normalization
CN109523526B (en) * 2018-11-08 2021-10-22 腾讯科技（深圳）有限公司 Tissue nodule detection and its model training method, device, equipment and system
US11222210B2 (en) * 2018-11-13 2022-01-11 Nec Corporation Attention and warping based domain adaptation for videos
EP3654245A1 (en) * 2018-11-13 2020-05-20 Nokia Technologies Oy Method and apparatus for processing sensor data
CN109523532B (en) * 2018-11-13 2022-05-03 腾讯医疗健康(深圳)有限公司 Image processing method, image processing device, computer readable medium and electronic equipment
US11188795B1 (en) * 2018-11-14 2021-11-30 Apple Inc. Domain adaptation using probability distribution distance
KR102756358B1 (en) * 2018-12-18 2025-01-17 삼성전자주식회사 Detector, method of object detection, learning apparatus, and learning method for domain transformation
JP7230521B2 (en) * 2019-01-17 2023-03-01 富士通株式会社 LEARNING METHOD, LEARNING PROGRAM AND LEARNING DEVICE
JP7172677B2 (en) * 2019-02-05 2022-11-16 富士通株式会社 LEARNING METHOD, LEARNING PROGRAM AND LEARNING DEVICE
US11120526B1 (en) 2019-04-05 2021-09-14 Snap Inc. Deep feature generative adversarial neural networks
CN110197183B (en) * 2019-04-17 2022-10-04 深圳大学 Image blind denoising method and device, computer equipment and storage medium
JP7276449B2 (en) * 2019-06-18 2023-05-18 日本電信電話株式会社 Generalized data generation device, estimation device, generalized data generation method, estimation method, generalized data generation program, and estimation program
EP4000009B1 (en) * 2019-07-17 2025-03-19 DeepMind Technologies Limited Training a neural network to control an agent using task-relevant adversarial imitation learning
EP3767536B1 (en) * 2019-07-17 2025-02-19 Naver Corporation Latent code for unsupervised domain adaptation
CN113994384B (en) 2019-09-11 2025-04-08 谷歌有限责任公司 Image coloring using machine learning
CN110689558B (en) * 2019-09-30 2022-07-22 清华大学 Multi-sensor image enhancement method and device
JP7166738B2 (en) * 2019-10-15 2022-11-08 Kddi株式会社 Information processing device, information processing method, and program
KR102345996B1 (en) * 2019-10-29 2021-12-30 연세대학교 산학협력단 Method and Device for Detecting Moving Object
US12236349B2 (en) * 2019-11-14 2025-02-25 Qualcomm Incorporated Guided training of machine learning models with convolution layer feature data fusion
US12266144B2 (en) * 2019-11-20 2025-04-01 Nvidia Corporation Training and inferencing using a neural network to predict orientations of objects in images
KR102869896B1 (en) * 2019-12-12 2025-10-14 삼성전자주식회사 Operating apparatus, operating method, and training method of network based on domain adaptation
CN111340819B (en) * 2020-02-10 2023-09-12 腾讯科技（深圳）有限公司 Image segmentation method, device and storage medium
US11468294B2 (en) * 2020-02-21 2022-10-11 Adobe Inc. Projecting images to a generative model based on gradient-free latent vector determination
CN113496237B (en) * 2020-03-20 2024-05-24 商汤集团有限公司 Domain adaptive neural network training and traffic environment image processing method and device
KR102703808B1 (en) 2020-04-21 2024-09-06 한국전자통신연구원 Apparatus and method for analyzing video image data and intelligent video apparatus having the same
CN111524521B (en) * 2020-04-22 2023-08-08 北京小米松果电子有限公司 Voiceprint extraction model training method, voiceprint recognition method, voiceprint extraction model training device and voiceprint recognition device
GB202016824D0 (en) 2020-10-23 2020-12-09 Deep Render Ltd DR big book 3
JP7391784B2 (en) 2020-07-06 2023-12-05 株式会社東芝 Information processing device, information processing method and program
EP3958167B1 (en) * 2020-08-21 2024-03-20 Toyota Jidosha Kabushiki Kaisha A method for training a neural network to deliver the viewpoints of objects using unlabeled pairs of images, and the corresponding system
CN112270208B (en) * 2020-09-28 2023-07-07 天津大学 Cross data domain micro-expression classification method based on feature subspace separation
CN112734049B (en) * 2020-11-23 2025-04-04 西湖大学 A multi-initialization meta-learning framework and method based on domain adaptation
KR102595207B1 (en) * 2021-01-11 2023-10-26 연세대학교 산학협력단 Apparatus and method for solving class imbalance problem of domain adaptation using content transfer
CN116888639B (en) 2021-01-18 2026-01-23 美国西门子医疗系统股份有限公司 Visual interpretation of classifications
CN114820414B (en) * 2021-01-22 2025-10-31 富士通株式会社 Neural network for field adaptation and training method thereof
CN112767368A (en) * 2021-01-25 2021-05-07 京科互联科技(山东)有限公司 Cross-domain medical image classification system and method based on domain self-adaption
CN112946567B (en) * 2021-01-26 2023-05-02 西安电子科技大学 Moving target fingerprint indoor positioning method based on domain antagonism neural network
US12373992B2 (en) * 2021-04-14 2025-07-29 Tencent America LLC Adaptive neural image compression with smooth quality control by meta-learning
KR20220159104A (en) * 2021-05-25 2022-12-02 삼성에스디에스 주식회사 Apparatus for Deep fake image discrimination and learning method thereof
KR102641500B1 (en) * 2021-10-06 2024-02-27 서울대학교산학협력단 Apparatus and method for unsupervised domain adaptation
US12229672B2 (en) * 2021-10-21 2025-02-18 EMC IP Holding Company LLC Detecting domain changes with domain classifiers in autonomous vehicles
KR102823815B1 (en) * 2021-12-15 2025-06-20 연세대학교 산학협력단 Semantic Segmentation Apparatus and Method using Memory
US12361614B2 (en) 2021-12-30 2025-07-15 Snap Inc. Protecting image features in stylized representations of a source image
CN114549987A (en) * 2022-02-24 2022-05-27 中山大学 Image processing method and image processing device based on multiple tasks
JP7814228B2 (en) * 2022-04-19 2026-02-16 株式会社東芝 Data processing device, data processing system and data processing method
KR102567128B1 (en) * 2023-01-30 2023-08-17 인빅 주식회사 Enhanced adversarial attention networks system and image generation method using the same
CN117078983B (en) * 2023-10-16 2023-12-29 安徽启新明智科技有限公司 Image matching method, device and equipment

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9158971B2 (en) * 2014-03-03 2015-10-13 Xerox Corporation Self-learning object detectors for unlabeled videos using multi-task learning
US10181098B2 (en) * 2014-06-06 2019-01-15 Google Llc Generating representations of input sequences using neural networks
US9922272B2 (en) 2014-09-25 2018-03-20 Siemens Healthcare Gmbh Deep similarity learning for multimodal medical images
CN104318214B (en) * 2014-10-27 2017-09-15 中国科学院自动化研究所 A kind of cross-view face identification method shifted based on structuring dictionary domain
CN104933722B (en) * 2015-06-29 2017-07-11 电子科技大学 A kind of method for detecting image edge based on Spiking convolutional neural networks models
CN105719000B (en) * 2016-01-21 2018-02-16 广西师范大学 A kind of neuron hardware unit and the method with this unit simulation impulsive neural networks

Also Published As

Publication number Publication date
EP3475879B1 (en) 2025-03-19
KR102318772B1 (en) 2021-10-28
US11361531B2 (en) 2022-06-14
US10970589B2 (en) 2021-04-06
US20190180136A1 (en) 2019-06-13
WO2018020277A1 (en) 2018-02-01
CN109643383A (en) 2019-04-16
CN109643383B (en) 2023-09-01
KR20190031318A (en) 2019-03-25
JP2019523504A (en) 2019-08-22
JP6771645B2 (en) 2020-10-21
US20210224573A1 (en) 2021-07-22

Similar Documents

Publication Publication Date Title
US11361531B2 (en) Domain separation neural networks
US12271823B2 (en) Training machine learning models by determining update rules using neural networks
US10991074B2 (en) Transforming source domain images into target domain images
US11775804B2 (en) Progressive neural networks
US12544913B2 (en) Domain adaptation using simulation to simulation transfer
CN111279362B (en) capsule neural network
US10748029B2 (en) Spatial transformer modules
US9224068B1 (en) Identifying objects in images
US11697205B2 (en) Determining control policies for robots with noise-tolerant structured exploration
CN110799992A (en) Using simulation and domain adaptation for robotic control
US20250182439A1 (en) Unsupervised learning of object keypoint locations in images through temporal transport or spatio-temporal transport
US11403807B2 (en) Learning hybrid (surface-based and volume-based) shape representation
US11453121B2 (en) Mirror loss neural networks

Legal Events

Date Code Title Description
STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: UNKNOWN

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE

PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE

17P Request for examination filed

Effective date: 20190128

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

AX Request for extension of the european patent

Extension state: BA ME

DAV Request for validation of the european patent (deleted)
DAX Request for extension of the european patent (deleted)
STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: EXAMINATION IS IN PROGRESS

17Q First examination report despatched

Effective date: 20210526

REG Reference to a national code

Ref country code: DE

Ref legal event code: R079

Free format text: PREVIOUS MAIN CLASS: G06K0009460000

Ipc: G06V0010770000

Ref country code: DE

Ref legal event code: R079

Ref document number: 602016091612

Country of ref document: DE

Free format text: PREVIOUS MAIN CLASS: G06K0009460000

Ipc: G06V0010770000

GRAP Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOSNIGR1

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: GRANT OF PATENT IS INTENDED

RIC1 Information provided on ipc code assigned before grant

Ipc: G06N 3/045 20230101ALI20240924BHEP

Ipc: G06N 3/084 20230101ALI20240924BHEP

Ipc: G06V 10/44 20220101ALI20240924BHEP

Ipc: G06V 10/72 20220101ALI20240924BHEP

Ipc: G06V 10/77 20220101AFI20240924BHEP

INTG Intention to grant announced

Effective date: 20241011

GRAS Grant fee paid

Free format text: ORIGINAL CODE: EPIDOSNIGR3

GRAA (expected) grant

Free format text: ORIGINAL CODE: 0009210

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE PATENT HAS BEEN GRANTED

AK Designated contracting states

Kind code of ref document: B1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

REG Reference to a national code

Ref country code: GB

Ref legal event code: FG4D

REG Reference to a national code

Ref country code: CH

Ref legal event code: EP

REG Reference to a national code

Ref country code: DE

Ref legal event code: R096

Ref document number: 602016091612

Country of ref document: DE

P01 Opt-out of the competence of the unified patent court (upc) registered

Free format text: CASE NUMBER: APP_10733/2025

Effective date: 20250304

REG Reference to a national code

Ref country code: IE

Ref legal event code: FG4D

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: RS

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250619

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: FI

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

REG Reference to a national code

Ref country code: LT

Ref legal event code: MG9D

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: NO

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250619

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: HR

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: LV

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: GR

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250620

Ref country code: BG

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

REG Reference to a national code

Ref country code: NL

Ref legal event code: MP

Effective date: 20250319

REG Reference to a national code

Ref country code: AT

Ref legal event code: MK05

Ref document number: 1777613

Country of ref document: AT

Kind code of ref document: T

Effective date: 20250319

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: NL

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: SE

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: SM

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: ES

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

Ref country code: PT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250721

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: DE

Payment date: 20250729

Year of fee payment: 10

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: PL

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

Ref country code: IT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: GB

Payment date: 20250728

Year of fee payment: 10

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: AT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: FR

Payment date: 20250728

Year of fee payment: 10

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: CZ

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

Ref country code: EE

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: RO

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: SK

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: IS

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250719

REG Reference to a national code

Ref country code: DE

Ref legal event code: R097

Ref document number: 602016091612

Country of ref document: DE

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: DK

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20250319

PLBE No opposition filed within time limit

Free format text: ORIGINAL CODE: 0009261

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT

REG Reference to a national code

Ref country code: CH

Ref legal event code: L10

Free format text: ST27 STATUS EVENT CODE: U-0-0-L10-L00 (AS PROVIDED BY THE NATIONAL OFFICE)

Effective date: 20260128

REG Reference to a national code

Ref country code: CH

Ref legal event code: H13

Free format text: ST27 STATUS EVENT CODE: U-0-0-H10-H13 (AS PROVIDED BY THE NATIONAL OFFICE)

Effective date: 20260224

26N No opposition filed

Effective date: 20251222

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: LU

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20250728

REG Reference to a national code

Ref country code: BE

Ref legal event code: MM

Effective date: 20250731

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: BE

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20250731

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: CH

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20250731